July 16, 2026. Two small lines in OpenAI's ChatGPT release notes this week change how seriously teams should treat ChatGPT configuration. On July 15, OpenAI raised the custom instructions limit from 1,500 to 5,000 characters for users on paid plans including Plus, Pro, Business, Enterprise and Education. A day earlier, on July 14, it rolled out unified search across chats, projects, images and documents on web, iOS and Android, on all plans globally. Release-note aggregators tracking OpenAI's feed logged both the same day.
What shipped
- Custom instructions at 5,000 characters (July 15). More than triple the previous 1,500-character ceiling, for paid plans. Custom instructions apply to every new conversation automatically.
- Unified search (July 14). One search box in the sidebar now finds past chats, projects, images and documents, with filters by content type, and opens the result directly. All plans, rolling out globally.
Why 5,000 characters is a real threshold
At 1,500 characters, custom instructions held a persona sketch. At 5,000, they hold an operating manual: brand voice rules, banned claims and required disclaimers, formatting standards for deliverables, escalation rules for uncertain answers, and the context about your business that every response should assume. That is not personalization. That is policy, and it now fits in the box on every seat your team runs. The change lands a week after the July 9 wave that shipped ChatGPT Work and the new desktop app, and it quietly completes the setup: agents to do the work, and enough instruction space to tell them how your company does it.
What it means for operators
- Write one canonical instruction block per team. Draft it in a shared doc, version it, and have every seat paste the same block. Consistent instructions are the cheapest quality control AI offers, and at 5,000 characters there is no longer an excuse for each person improvising their own.
- Spend the budget on constraints, not compliments. Output format rules, claim restrictions, citation requirements and escalation behavior beat flattering persona text. Treat it like a standing brief for a contractor.
- Unified search makes ChatGPT a work archive, so govern it like one. Everything typed into chats is now trivially findable by the account holder. That is a productivity win and a reminder: client data and credentials that should not live in a chat log are now searchable chat log. Set rules for what goes in.
Standardizing AI configuration across a company is a smaller version of what we build for clients at our AI automation agency, and rolling AI into daily operations with guardrails is the core of our AI automation service. The tools keep getting cheaper to run well. The gap is whether anyone writes the manual.
Frequently Asked Questions
Two things, per OpenAI's official release notes. On July 14, 2026, ChatGPT gained unified search across chats, projects, images and documents on web, iOS and Android for all plans. On July 15, the custom instructions character limit rose from 1,500 to 5,000 characters for paid plans including Plus, Pro, Business, Enterprise and Education.
Treat it as a standing brief: who the company is and who it serves, brand voice rules, banned claims and required disclaimers, default output formats for common deliverables, citation expectations, and an escalation rule telling the model to flag uncertainty instead of guessing. Draft it once centrally, version it, and use the same block on every seat.
No. Custom instructions apply to new conversations after they are saved. Existing chats keep the context they were started with, so roll out an updated instruction block and start fresh threads for work where the new rules matter.
It raises the stakes. Content typed into ChatGPT is now easy to retrieve through one search box across chats, projects and uploaded documents. That is useful for finding past work and a good prompt to set team rules about what belongs in chats at all, especially client identifiers, credentials and regulated data.